Multiple Oracle Algorithm to Solve Continuous Games

Continuous games are multiplayer in which strategy sets compact and utility functions continuous. These typically have a highly complicated structure of Nash equilibria, numerical methods for the equilibrium computation known only particular classes continuous games, such as two-player polynomial or pure equilibria guaranteed to exist. This contribution focuses on approximation mixed whole class general-sum games. We vastly extend scope applicability double oracle algorithm, initially designed proved converge zero-sum Specifically, we propose an iterative generation technique, splits original problem into master with finite subset strategies being considered, subproblem finds best response each player. simple method is recover approximate finitely many iterations. Further, argue that Wasserstein distance (the earth mover’s distance) right metric space our purposes. Our main result convergence this algorithm game. The experiments show performance several including randomly generated examples.